In [2]:
#Approach 1: Using Word Cloud
!pip install wordcloud
In [1]:
# Importing the necessery modules
from wordcloud import WordCloud, STOPWORDS
import matplotlib.pyplot as plt
import csv
# file object is created
file_ob = open(r"C:\\Users\YVKCH\Desktop\Python Codes\\amazon_reviews.csv")
stopwords = set(STOPWORDS)
stopwords1=list(stopwords)
print(stopwords1)
print(type(stopwords1))
stopwords2=["for", "account", "not","on","FOR", "ACCOUNT","NOT","ON","is","This","For","The","the","hi","It","it","and"
, "And","this","to","It is"]
print(type(stopwords2))
stopwords_final=stopwords1+stopwords2
print(stopwords_final)
# reader object is created
reader_ob = csv.reader(file_ob)
# contents of reader object is stored .
# data is stored in list of list format.
reader_contents = list(reader_ob)
# empty string is declare
text = ""
# iterating through list of rows
for row in reader_contents :
# iterating through words in the row
for word in row :
# concatenate the words
text = text + " " + word
wordcloud = WordCloud(width=1000, height=1000,
stopwords=stopwords_final, background_color='white').generate(text) #stopwords_final is enetered into wordcloud
# plot the WordCloud image
plt.figure(figsize = (8, 8), facecolor = None)
plt.imshow(wordcloud)
plt.axis("off")
plt.tight_layout(pad = 0)
plt.show()
In [ ]:
#Insights from the above WordCloud
#1)Most customers had positive reviews about the products (identified via words- Great, Loves, easy to use)
#2)It looked like customers presented Electronics as gifts to son/daughter (many positive reviews expressed by customers are tied to son/daughter)
In [3]:
#Approach 2: TextHero
!pip install texthero
In [4]:
import texthero as hero
import pandas as pd
import csv
import matplotlib as plt
In [6]:
#Importing only the Product Name and the textual reviews of the customer into the Dataframe for Text Mining
#I have encountered utf-8 encoding error when I read the file. Using encoding='cp1252' resolved the error
df = pd.read_csv('Amazon_Textual_Reviews_Electronics.csv',encoding='cp1252')
In [7]:
#Sample 5 records from the file
df.head(5)
Out[7]:
In [8]:
#Step 1: Cleaning the textual reviews with hero.clean
df['clean_text'] = hero.clean(df['Review'])
In [9]:
#Step 2: This step performs the same cleaning as Step 1. However, it takes into account only certain pre-processing cleaning steps as indicated "custom"
from texthero import preprocessing
custom = [preprocessing.fillna,
preprocessing.lowercase,
preprocessing.remove_whitespace,
preprocessing.remove_digits,
preprocessing.remove_punctuation,
preprocessing.remove_diacritics,
preprocessing.remove_stopwords,
preprocessing.stem]
df['clean_text'] = hero.clean(df['Review'])
In [10]:
#Step 3: Adding Term Frequency vector and PCA values into the data
df['tfidf_clean_text'] = hero.tfidf(df['clean_text'])
df['pca_tfidf_clean_text'] = hero.pca(df['tfidf_clean_text'])
In [11]:
#Step 4: This step peforms the same function as Step1 and Step 3 combined together with the help of pipe function in pandas
df['pca'] = (
df['clean_text']
.pipe(hero.clean)
.pipe(hero.tfidf)
.pipe(hero.pca)
)
In [12]:
df.head()
Out[12]:
In [14]:
#Plot 1
#Scatter plot of PCA vector values
hero.scatterplot(df, col='pca', color='Product', title="PCA Amazon Product Reviews")
In [15]:
#Pulling the top words by product
NUM_TOP_WORDS = 10
df.groupby('Product')['clean_text'].apply(lambda x: hero.top_words(x)[:NUM_TOP_WORDS])
#You can see that the top 10 words are pulled for each different product
#If you take the first product, All-New Fire HD 8 Tablet- you can see that customers feel great about the product and its easy to use
Out[15]:
In [16]:
#Plot 2: Plotting K-means
#pipe is the Pandas function used when chaining together functions
df['tfidf2'] = (
df['Review']
.pipe(hero.clean)
.pipe(hero.tfidf)
)
df['kmeans_labels'] = (
df['tfidf2']
.pipe(hero.kmeans, n_clusters=5) #Defining 5 clusters (Forcing the algorithm to use only 5 clusters)
.astype(str)
)
df['pca_k'] = df['tfidf2'].pipe(hero.pca)
hero.scatterplot(df, 'pca_k', color='kmeans_labels', title="K-means Plot for Amazon Customer Reviews")
In [ ]: